generate-reliability-skill — Claude Code skill: installation
The generate-reliability-skill skill for Claude Code and Codex creates a SKILL.md with reliability rules for agent SDKs found in a repository. It activates before agent code is written or changed.
Apache-2.0 ★ 62 updated 2026-11-15
/plugin marketplace add trustabl/agent-reliability-skills
/plugin install agent-reliability-skills@agent-reliability-skillsWhat generate-reliability-skill does
The skill checks dependency manifests, identifies the agent SDKs a project uses, then asks Trustabl Forge to produce a matching SKILL.md. The generated file goes to .claude/skills/trustabl-pre-coding/SKILL.md. An existing file is replaced only when it begins with a Forge stamp; a hand-written skill without that stamp is left untouched.
Trustabl Analyzer 0.1.14 or later is required. If trustabl is not on PATH, the setup script downloads a pinned release and verifies it against checksums.txt; it uses Bash, curl, and an archive extractor. Forge fetches Trustabl’s rules on its first run and uses a cache afterward. It makes no model call and needs no API key. If manifest scanning finds no SDKs, --policy can add SDK categories explicitly.
What it helps with
- detect agent SDKs in repository dependency manifests
- generate reliability rules for the SDKs a project uses
- save a pre-coding skill for agent development
Example prompts for the agent
- Create pre-coding reliability rules for the agent code in this repository.
- Generate a Trustabl Forge skill for the SDKs used in this project.
- Add rules for
mcpandopenai_sdk.
How to install generate-reliability-skill
1Plugin marketplace
/plugin marketplace add trustabl/agent-reliability-skills/plugin install agent-reliability-skills@agent-reliability-skillsThis installs the whole agent-reliability-skills plugin with all its skills
2Skills CLI (Vercel)
npx skills add trustabl/agent-reliability-skills --skill generate-reliability-skill -a claude-code3Manually: put the files in place
~/.claude/skills/generate-reliability-skill/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/generate-reliability-skill/
1Skills CLI (Vercel)
npx skills add trustabl/agent-reliability-skills --skill generate-reliability-skill -a codex2Manually: put the files in place
.agents/skills/generate-reliability-skill/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add trustabl/agent-reliability-skills --skill generate-reliability-skill -a cursor2Manually: put the files in place
.cursor/skills/generate-reliability-skill/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install trustabl/agent-reliability-skills claude-plugin/skills/generate-reliability-skill2Manually: put the files in place
.github/skills/generate-reliability-skill/SKILL.md1Skills CLI (Vercel)
npx skills add trustabl/agent-reliability-skills --skill generate-reliability-skill -a gemini-cli2Platform CLI
gemini skills install https://github.com/trustabl/agent-reliability-skills --path claude-plugin/skills/generate-reliability-skillManually: put the files in place
.devin/skills/generate-reliability-skill/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about generate-reliability-skill
How do I install generate-reliability-skill in Claude Code?
Run these commands in order: /plugin marketplace add trustabl/agent-reliability-skills, then /plugin install agent-reliability-skills@agent-reliability-skills.
What does generate-reliability-skill do?
It scans repository dependency manifests for agent SDKs and generates a SKILL.md with Trustabl Forge rules for the SDKs it finds.
What does generate-reliability-skill need to run?
It requires Trustabl Analyzer 0.1.14 or later. When trustabl is not on PATH, the download script uses Bash, curl, and an archive extractor. No API key is needed.
Source text
SKILL.md · Apache-2.0
---
name: generate-reliability-skill
description: >-
Use before writing or changing AI agent code in a repository (Claude Agent SDK,
OpenAI Agents SDK, Google ADK, MCP, LangChain/LangGraph, CrewAI, Pydantic AI,
Vercel AI SDK, AutoGen), or when the user asks for Trustabl Forge or a
pre-coding reliability skill. Generates a SKILL.md from Trustabl's detection
rules for the agent SDKs the repository uses, so you apply them while you code.
---
# Generate a pre-coding reliability skill with Trustabl Forge
Forge reads the repository's dependency manifests (`package.json`,
`pyproject.toml`, `requirements.txt`, `go.mod` and others), finds the agent SDKs
it uses, and prints a `SKILL.md` built from Trustabl's detection rules for exactly
those SDKs. It makes no model call and needs no API key.
## Run it
Forge is part of the open-source Trustabl analyzer (`trustabl forge`, 0.1.14 or later).
The plugin ships no scripts. Run this block with the Bash tool: it downloads the pinned release from GitHub,
checks it against the release's `checksums.txt` (a mismatch installs nothing), keeps it in `~/.trustabl/cli/`, and
prints `TRUSTABL_BIN=<path>`. Later runs only check the version, so run it at the start of each session that uses
Forge. Tell the user what it downloads the first time. If the user already has `trustabl` 0.1.14 or later on `PATH` (`trustabl version`), use that instead.
```bash
V=0.1.16; D="$HOME/.trustabl/cli/v$V"
case "$(uname -s)" in MINGW*|MSYS*|CYGWIN*) OS=windows; EXT=zip; EXE=trustabl.exe ;; Darwin) OS=darwin; EXT=tar.gz; EXE=trustabl ;; *) OS=linux; EXT=tar.gz; EXE=trustabl ;; esac
case "$(uname -m)" in arm64|aarch64) ARCH=arm64 ;; *) ARCH=amd64 ;; esac; [ "$OS" = windows ] && ARCH=amd64
TRUSTABL_BIN="$D/$EXE"
if ! "$TRUSTABL_BIN" version 2>/dev/null | head -n1 | grep -q "$V\$"; then
A="trustabl_${V}_${OS}_${ARCH}.$EXT"; U="https://github.com/trustabl/agent-reliability-analyzer/releases/download/v$V"; T="$(mktemp -d)"
curl -fsSL "$U/$A" -o "$T/$A" && curl -fsSL "$U/checksums.txt" -o "$T/checksums.txt" &&
SUM=sha256sum; command -v sha256sum >/dev/null || SUM="shasum -a 256"
(cd "$T" && L="$(grep " $A\$" checksums.txt)" && [ -n "$L" ] && echo "$L" | $SUM -c -) &&
mkdir -p "$D" && if [ "$EXT" = zip ]; then unzip -oq "$T/$A" -d "$D"; else tar -xzf "$T/$A" -C "$D"; fi &&
chmod +x "$TRUSTABL_BIN"; rm -rf "$T"
fi
if "$TRUSTABL_BIN" version 2>/dev/null | head -n1 | grep -q "$V\$"; then echo "TRUSTABL_BIN=$TRUSTABL_BIN"; else echo "install failed: see the message above"; fi
```
Shell state does not carry between Bash calls: start each later command with `TRUSTABL_BIN="<the printed path>";`.
If the block prints `install failed`, show the user the message above it and stop.
```bash
"$TRUSTABL_BIN" forge <repo> # detected SDKs only
"$TRUSTABL_BIN" forge <repo> --policy mcp,openai_sdk # add SDK categories not yet in a manifest
"$TRUSTABL_BIN" forge <repo> --lang python,typescript # add languages: python, typescript, go
```
`<repo>` is `.` unless the user named another directory. The SKILL.md goes to
stdout; any note or warning goes to stderr. The first run also fetches
Trustabl's rules; after that it works from cache.
**No SDK detected:** exit code 1 with `no SDKs detected` on stderr means no
supported SDK is declared in the manifests. If the repository does use one (for
example declared only in a subfolder's manifest), run again with `--policy`
naming it, and say so.
## Save it
Save the output to `.claude/skills/trustabl-pre-coding/SKILL.md`:
- If no file is there, write it.
- If one is there, replace it **only if it starts with a Forge stamp**
(`<!-- generated: `); a file without the stamp is a hand-written skill: never
overwrite it, tell the user instead and offer another path.
The stamp line reads `<!-- generated: <date> | rules: <version> | schema: <n> |
sdks: <list> | template: <n> -->`. Report the SDKs it covers and the rules
version, then follow the skill's rules while you write code.
© Trustabl · Apache-2.0 · trustabl/agent-reliability-skills